[cairo] Cairo performance comparison, 770 / N800 / PXA-320

[Daniel Amelang]

On 1/13/07, Kalle Vahlman <kalle.vahlman at gmail.com> wrote:
> Hi!
>
> We have all sorts of funny hardware at the office, so I thought I'd
> make a quick run of cairo-perf with the Cairo 1.3.10 snapshot and see
> how they relate to each other.
> ...
> N800 is naturally faster than 770, but I didn't expect the xlib
> backend to have so big differences between the two.

Yea, if it were just hardware, I would think that our image surface
would be getting about the same boost as the X side. Clearly it's not.

So, the X versions are clearly different. Hmmm....6.6.3 vs. 1.1.99.3 I
can't find much info about what that 6.6.3 version is all about. Plus,
who knows what xserver-xomap specific optimizations Nokia has been up
to. You know, the OMAP 2420 has ARM v6 SIMD and a PowerVR (!), so
maybe the xomap on the N800 takes advantage of some of that extra
power. If not, I may have found my future class project for my
embedded software class :)

>
> For the cairo audience there's the question of the tessellation
> process, can it really be so fast on the PXA-320 or is there a bug
> somewhere that twists the results? What could be so good in PXA-320
> (or not-good on the other devices) that the results are so drastic?

If I remember correctly, the tessellator performs a lot of integer
div/mod, which is dreadfully slow on OMAP ARM due to lack of hardware
support. Plus, much of the arithmetic is performed on 64-bit integers,
which is also much slower than 32-bit on OMAP IIRC.

So, if the PXA-320 ARM doesn't have either of the two limitations (I
have no idea, but it is a very different ARM implementation), you
should expect to see a healthy speedup. Although 80x is pretty
drastic.

Side note: the N800 has VFP, so we should have seen a large
performance improvement for FP-heavy cairo perf tests, like the radial
stuff. But I'm only seeing ~2x improvement. Anyone know what's going
here?

Another side note: did you run cairo-test on these platforms? I've
been uneasy about cairo on ARM since that fellow last month reported
the strange diagonal line problem that we never resolved.

Dan